Low-altitude economy unmanned aerial vehicle data processing method and system

Abstract

The present application relates to target recognition technical field, specifically to low altitude economy unmanned plane data processing method and system, including the following steps: through multiple airborne cameras obtain target sequence and extract curvature to construct feature library, based on feature library threshold value judgment trigger scale correction output standardized data set, multi-view target area division sub-block extracts centroid coordinate analysis output positioning data, call positioning data normalization compensation and kalman filter output tracking trajectory, based on trajectory and task parameter threshold value judgment fusion detection trigger alarm planning output running state, the present application improves edge feature stability through multi-angle synchronous acquisition and curvature extraction, keeps multi-machine target contour consistent through cross-scale threshold value trigger distribution correction, combines centroid displacement vector statistics direction and amplitude through geometric division, introduces normalization and filter compensation to realize smooth positioning, obtains stable trajectory to improve collaborative identification and spatial positioning accuracy.

Description

Technical Field

[0001] This invention relates to the field of target recognition technology, and in particular to a data processing method and system for low-altitude economic unmanned aerial vehicles (UAVs). Background Technology

[0002] The field of target recognition technology belongs to the intersection of artificial intelligence and information processing. Its core matters include the perception, detection and recognition of target objects in different scenarios. This technology covers the entire process from the acquisition of perception information and feature extraction to target classification and labeling. It relies on the fusion of computer vision and sensors to realize the recognition and discrimination of static or dynamic targets. In low-altitude economic applications, it involves UAVs equipped with perception devices to collect and process multi-source data in low-altitude environments.

[0003] Traditional low-altitude economic UAV data processing methods refer to the processing of image or video data acquired by UAVs during flight missions through methods such as data framing, target contour extraction, and feature point matching. At the same time, continuous identification of the same target is achieved through the integration of time-series data. The methods used include edge detection-based image processing, target comparison based on color or texture features, time-series consistency verification based on multi-frame data, and auxiliary identification using flight trajectory parameters and environmental data.

[0004] Traditional methods rely on edge detection and feature comparison to obtain target information. When the lighting conditions are uneven or the background is cluttered, edge pixels are prone to breakage, resulting in insufficient integrity of target information. During the superposition of multiple frames of data, scale differences and target occlusion will further weaken the temporal continuity, thereby affecting the spatial positioning accuracy. If the environmental interference is strong during the trajectory calculation process, it is easy to cause fluctuations in the discrimination parameters, resulting in jumps or deviations in the tracking process. The recognition and positioning results are difficult to meet the stable requirements of complex low-altitude environments. Summary of the Invention

[0005] To address the technical problems of traditional methods that rely on edge detection and feature comparison to obtain target information, which are prone to edge pixel breakage under uneven lighting conditions or cluttered backgrounds, resulting in insufficient target information integrity, scale differences and target occlusion further weakening temporal continuity during multi-frame data overlay, thus affecting spatial positioning accuracy, and strong environmental interference during trajectory calculation, which can easily cause fluctuations in discrimination parameters, leading to jumps or deviations in the tracking process, and making it difficult for the recognition and positioning results to meet the stability requirements of complex low-altitude environments, this invention provides a data processing method and system for low-altitude economic unmanned aerial vehicles (UAVs).

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The data processing method for low-altitude economic unmanned aerial vehicles includes the following steps:

[0008] S1: Simultaneously acquire distributed target image sequences through multiple airborne cameras, perform parallel curvature extraction processing on target edge pixels in the distributed target image sequences, construct local tangent direction vectors, calculate principal curvature and secondary curvature values, and output a multi-camera target feature vector library;

[0009] S2: Based on the multi-machine target feature vector library, the rate of change of curvature values ​​under cross-machine and cross-scale conditions is judged by a threshold. When it exceeds the preset threshold range, a distributed scale correction mechanism is triggered. A Gaussian mixture model is used to perform multi-machine edge pixel collaborative relocation and output a standardized target dataset.

[0010] S3: Based on the standardized target dataset, construct a multi-view target region, divide the multi-view target region into multiple geometric sub-blocks and extract the centroid coordinates of the sub-blocks, calculate the centroid coordinate displacement vector, perform direction and amplitude statistical analysis, and output UAV cluster spatial positioning data.

[0011] S4: Call the spatial positioning data of the UAV cluster to perform weighted mean normalization operation, extract the correction vector value and perform parallel translation compensation on the centroid coordinates of the sub-blocks, use the Kalman filter algorithm for filtering processing, and output the low-altitude target tracking trajectory data.

[0012] As a further embodiment of the present invention, the multi-machine target feature vector library includes curvature distribution features, boundary direction features, and shape contour features; the standardized target dataset includes scale-uniform data, coordinate-normalized data, and pixel-consistent data; the UAV swarm spatial positioning data includes regional geometric features, centroid distribution features, and displacement vector features; and the low-altitude target tracking trajectory data includes direction sequence, amplitude sequence, and time sequence.

[0013] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0014] S101: Simultaneously acquire distributed target image sequences through multiple airborne cameras, perform point-by-point geometric curvature extraction on the edge pixels of the target region in the distributed target image sequence, and arrange the curvature values ​​in index order to generate a curvature sequence set;

[0015] S102: Call the curvature sequence set, vectorize the tangential direction of adjacent pixels, numerically calculate the angle difference and direction change between the tangential vectors, and encode the difference results by index to obtain the tangential direction vector set;

[0016] S103: Based on the tangent direction vector set, solve the principal curvature and secondary curvature values ​​of multiple pixels point by point, and then combine the numerical results according to the feature dimension to establish a multi-machine target feature vector library.

[0017] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0018] S201: Based on the feature vector sequence recorded in the multi-machine target feature vector library, extract the curvature values ​​point by point under cross-machine and cross-scale conditions, compare the rate of change with a preset curvature threshold, and mark when the rate of change exceeds the threshold to generate a curvature anomaly marker set.

[0019] S202: Call the curvature anomaly marker set, retrieve the scale parameters of the marker positions one by one, calculate the difference between the multi-scale parameters and the unified reference scale, and then compare it with the correction threshold. Record and merge the indexes that exceed the threshold to obtain the scale correction index set.

[0020] S203: Based on the scale-corrected index set, perform probability density distribution estimation on the coordinate vectors of the multi-machine edge pixels corresponding to the index, use a Gaussian mixture model to fit the density and perform collaborative weighted localization, and aggregate the coordinate vectors to obtain a standardized target dataset.

[0021] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0022] S301: Based on the standardized target dataset, construct a multi-view target region, perform geometric subdivision operation on the target region and aggregate the boundary point coordinates to form a spatial range, perform average operation on the coordinate point set within the range, and generate a sub-block centroid coordinate set.

[0023] S302: Call the sub-block centroid coordinate set, perform difference operation on the same sub-block centroid coordinates in adjacent observation times, and reassemble the difference results into an ordered vector sequence after decomposing the three-dimensional axial components to obtain the centroid coordinate displacement vector set.

[0024] S303: Based on the centroid coordinate displacement vector set, perform aggregate statistics on the direction angle and amplitude parameters of the vectors, and arrange the statistical results in order into a matrix structure to obtain the spatial positioning data of the UAV cluster.

[0025] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0026] S401: Based on the spatial positioning data of the UAV cluster, the three-dimensional parameters of the coordinate points in the data are numerically weighted according to the weight coefficients. The weighted coordinate set is compared with the sum parameter matrix and then a normalization operation is performed to generate a correction vector value.

[0027] S402: Call the corrected vector value, perform translation compensation calculations on the horizontal and vertical parameters in the centroid coordinate matrix of the sub-block respectively, update the calculated coordinate points in parallel, and obtain the translation compensation coordinate set;

[0028] S403: Based on the translation compensation coordinate set, the predicted and observed values ​​of the time series coordinate points are differentially analyzed. The Kalman filter algorithm is used to input the differential result into the state transition matrix and is combined with the observation residual vector for correction to obtain low-altitude target tracking trajectory data.

[0029] As a further aspect of the present invention, the low-altitude economic unmanned aerial vehicle (UAV) data processing method further includes the following steps:

[0030] S5: Based on the low-altitude target tracking trajectory data and preset flight mission parameters, a matching degree threshold is determined, multi-aircraft trajectory data is fused and analyzed and anomalies are detected, and when the preset threshold is exceeded, an alarm mechanism is triggered and the flight path is replanned, and the low-altitude economic UAV operation status result is output.

[0031] The operational status results of the low-altitude economic UAV include mission matching degree, abnormal alarm information, and flight path planning scheme.

[0032] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0033] S501: Based on the low-altitude target tracking trajectory data and preset flight mission parameters, calculate the Euclidean distance between the trajectory data position coordinates and mission track points one by one, compare the difference with the matching degree threshold, record the coordinate index below the threshold and perform time serialization encoding to generate a trajectory matching index sequence.

[0034] S502: Call the trajectory matching index sequence, extract the corresponding velocity vector from the multi-machine trajectory dataset, split the velocity vectors of the same time slice into direction angle and magnitude and calculate the difference, compare the difference with the anomaly detection threshold, mark the time slice index that exceeds the threshold, and obtain the abnormal time slice set.

[0035] S503: Based on the set of abnormal time slices, call the task parameter waypoints, reassemble the reallocated velocity vectors and position coordinates to form a new waypoint sequence, and synchronize it with the original flight state parameters in the time dimension to obtain the low-altitude economic UAV operation status result.

[0036] As a further aspect of the present invention, the matching degree threshold is determined based on the UAV navigation and positioning accuracy and the allowable deviation of trajectory tracking;

[0037] The anomaly detection threshold is determined based on the fluctuation range of the UAV's speed modulus and the deviation range of its azimuth angle.

[0038] Meanwhile, the present invention also provides a low-altitude economic unmanned aerial vehicle (UAV) data processing system, comprising:

[0039] The image curvature extraction module is configured to simultaneously acquire a distributed target image sequence through multiple airborne cameras, perform parallel curvature extraction processing on the target edge pixels in the distributed target image sequence, construct a local tangent direction vector, calculate the principal curvature value and the secondary curvature value, and output a multi-camera target feature vector library.

[0040] The multi-machine feature modeling module is configured to perform threshold judgment on the rate of change of curvature values ​​under cross-machine and cross-scale conditions based on the multi-machine target feature vector library. When the rate of change exceeds the preset threshold range, a distributed scale correction mechanism is triggered, and a Gaussian mixture model is used to perform multi-machine edge pixel collaborative relocation and output a standardized target dataset.

[0041] The target region segmentation module is configured to construct a multi-view target region based on the standardized target dataset, divide the multi-view target region into multiple geometric sub-blocks and extract the centroid coordinates of the sub-blocks, calculate the centroid coordinate displacement vector, perform direction and amplitude statistical analysis, and output UAV cluster spatial positioning data.

[0042] The spatial positioning compensation module is configured to call the spatial positioning data of the UAV cluster to perform weighted mean normalization operation, extract the correction vector value and perform parallel translation compensation on the centroid coordinates of the sub-block, use the Kalman filter algorithm for filtering, and output low-altitude target tracking trajectory data.

[0043] The trajectory fusion determination module is configured to determine the matching degree threshold based on the low-altitude target tracking trajectory data and preset flight mission parameters, perform fusion analysis and anomaly detection on multi-aircraft trajectory data, trigger an alarm mechanism and replan the flight path when the preset threshold is exceeded, and output the low-altitude economic UAV operation status result.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0045] In this invention, by simultaneously acquiring data from multiple angles and extracting curvature values, higher stability can be achieved in the expression of edge features. Combined with numerical threshold judgment under cross-scale conditions to trigger distributed correction, the target contours in multi-camera images remain consistent. Furthermore, geometric division is performed on a unified dataset, and statistical analysis is conducted using centroid displacement vectors to achieve accurate representation of direction and amplitude. On this basis, normalization and filtering methods are introduced for dynamic compensation, so that the positioning results remain smooth and continuous in complex environments. Finally, a stable and reliable target tracking trajectory is obtained, effectively improving the robustness and accuracy of cluster collaborative recognition and spatial positioning. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the steps of the present invention.

[0048] Figure 2 This is a detailed schematic diagram of S1 of the present invention.

[0049] Figure 3 This is a detailed schematic diagram of S2 in this invention.

[0050] Figure 4 This is a detailed schematic diagram of S3 of the present invention.

[0051] Figure 5 This is a detailed schematic diagram of S4 of the present invention.

[0052] Figure 6 This is a detailed schematic diagram of S5 of the present invention.

[0053] Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0054] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0055] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0056] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0057] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0058] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0059] Please see Figure 1 This invention provides a data processing method for low-altitude economic unmanned aerial vehicles (UAVs), comprising the following steps:

[0060] S1: Simultaneously acquire distributed target image sequences through multiple airborne cameras, perform parallel curvature extraction processing on target edge pixels in the distributed target image sequences, construct local tangent direction vectors, calculate principal curvature and secondary curvature values, and output a multi-camera target feature vector library;

[0061] S2: Based on the multi-machine target feature vector library, threshold judgment is made on the rate of change of curvature values ​​under cross-machine and cross-scale conditions. When it exceeds the preset threshold range, a distributed scale correction mechanism is triggered. A Gaussian mixture model is used to perform multi-machine edge pixel collaborative relocalization and output a standardized target dataset.

[0062] S3: Obtain a standardized target dataset to form a multi-view target region, divide the multi-view target region into multiple geometric sub-blocks and extract the centroid coordinates of the sub-blocks, calculate the centroid coordinate displacement vector, perform direction and amplitude statistical analysis, and output UAV cluster spatial positioning data.

[0063] S4: Call the spatial positioning data of the UAV cluster to perform weighted mean normalization operation, extract the correction vector value and perform parallel translation compensation on the centroid coordinates of the sub-blocks, use the Kalman filter algorithm for filtering, and output the low-altitude target tracking trajectory data.

[0064] S5: Based on the low-altitude target tracking trajectory data and preset flight mission parameters, the matching degree threshold is judged, the multi-aircraft trajectory data is fused and analyzed and anomalies are detected. When the preset threshold is exceeded, an alarm mechanism is triggered and the flight path is replanned, and the low-altitude economic UAV operation status result is output.

[0065] The multi-aircraft target feature vector library includes curvature distribution features, boundary direction features, and shape contour features. The standardized target dataset includes scale-uniform data, coordinate-normalized data, and pixel-consistent data. The UAV swarm spatial positioning data includes regional geometric features, centroid distribution features, and displacement vector features. The low-altitude target tracking trajectory data includes direction sequences, amplitude sequences, and time sequences. The low-altitude economic UAV operation status results include task matching degree, abnormal alarm information, and trajectory planning scheme.

[0066] Please see Figure 2 The specific steps of S1 are as follows:

[0067] S101: Simultaneously acquire distributed target image sequences through multiple airborne cameras, perform point-by-point geometric curvature extraction on the edge pixels of the target region in the distributed target image sequence, and arrange the curvature values ​​in index order to generate a curvature sequence set;

[0068] Using industrial cameras mounted on three drones, an image sequence containing the crack target was simultaneously acquired at a height of 10 meters above the bridge structure surface. First, the frame with timestamp T0 was extracted from the image sequence. The image data of one of the drones was processed, with a resolution of 4096×3000 pixels. The target area, i.e., the area containing the crack, was located by edge pixel localization. Specifically, for each pixel in the image, the sum of the squares of the grayscale differences between it and its eight neighboring pixels was calculated to obtain a gradient intensity value. For example, if pixel P0 at coordinates (100, 250) has a grayscale value of 128, its right neighbor P1 has a grayscale value of 150, and its lower neighbor P2 has a grayscale value of 135, then its horizontal gradient is approximately 150-128=22, and its vertical gradient is approximately 135-128=7. The gradient intensity of this point is approximately... After calculating the gradient intensity of each pixel, a gradient intensity threshold G is set. th To distinguish between edge and non-edge regions, this threshold is set based on the overall gradient intensity histogram of the image. The gradient intensity value is selected when the cumulative number of pixels with gradient intensity values ​​from high to low reaches 5% of the total number of pixels. Assuming this value is calculated to be 20.5, pixels with a gradient intensity greater than 20.5 are initially identified as edge points. Subsequently, connectivity analysis is performed on these edge points to remove isolated noise points and form continuous chains of edge pixels. For example, from edge point P... i (x i y i The process begins by searching for any remaining edge points within its 3×3 neighborhood. If any exist, these are connected, and the search continues until no more edge points can be connected, forming a complete sequence of crack edge contour pixels. Next, the geometric curvature is extracted point-by-point for each pixel in this sequence. For any pixel P in the sequence... i (x i y i Select the point P preceding it in the sequence. i-1 (x i-1 y i-1 ) and the next point P i+1 (x i+1 y i+1 Using these three points to estimate P i The curvature at that point, for example, for pixel P with index 10. 10Its coordinates are (150, 320), and its predecessor and successor point coordinates are P9(148, 318) and P... 11 (152, 321), first calculate the lengths of the three sides of the triangle formed by the three points:

[0069] ;

[0070] ;

[0071] ;

[0072] Next, calculate the area A of the triangle. Using Heron's formula, the semi-perimeter... ,area:

[0073] ;

[0074] Finally, point P 10 curvature value K 10 via 4A / (d) 9,10 ·d 10,11 ·d 9,11 The value is calculated as follows: This calculation process is repeated for each pixel on the edge contour, and the calculated curvature value K is then used. i The curvature values ​​are associated with their corresponding pixel index i, and finally arranged according to the original index order of the pixels in the contour sequence to generate a one-dimensional array, which is the curvature sequence set of the crack target.

[0075] S102: Call the curvature sequence set, vectorize the tangential direction of adjacent pixels, numerically calculate the angle difference and direction change between the tangential vectors, and encode the difference results by index to obtain the tangential direction vector set;

[0076] The curvature sequence set and the corresponding edge pixel coordinate sequence are called, and the tangential direction formed by adjacent pixels in the sequence is vectorized. Specifically, for any non-endpoint pixel P in the edge pixel sequence... i (x i y i ), its tangential direction vector Through its preceding pixel P i-1 (x i-1 y i-1 ) and subsequent pixel P i+1 (x i+1 y i+1 The coordinate difference is used to determine the value, i.e. For example, using pixel P 10(150, 320), its preorder and postorder points are P9(148, 318) and P, respectively. 11 (152, 321), then at P 10 Tangential vector at point The calculation is (152-148, 321-318) = (4, 3). Similarly, for P... 11 (152, 321), assuming its successor point is P. 12 (154, 323), then its tangential vector The calculation is (154-150, 323-320) = (4, 3). Next, the angle difference between the tangent vectors of two consecutive pixels is numerically calculated. This angle difference reflects the local curvature of the edge contour. The angle θ i By calculating adjacent vectors and It is obtained by taking the dot product and the modulus, and the calculation process is as follows: Continuing with the previous example, Assume P 12 The subsequent point is P 13 (155, 326), then Now calculate and The angle difference Δθ between them 11 dot product , module length , module length Then the cosine of the included angle is The corresponding angle difference Δθ 11 The angle is approximately 22.17 degrees. Meanwhile, the change in direction is defined as the magnitude of the difference vector between these two vectors, i.e. ;

[0077] Then, the calculated angle difference Δθ i Accumulated encoding is performed based on the index order of the pixels. The initial accumulated value is set to A0=0. The first encoded value is A1=A0+Δθ0, the second is A2=A1+Δθ1, and so on. If the difference between the first three included angles is 2.5 degrees, 4.1 degrees, and 22.17 degrees, the corresponding accumulated encoding sequence is 2.5, 6.6, 28.77, ... Each value in this accumulated encoding sequence represents the total angle change from the starting point of the contour to the current point. This accumulated encoding sequence, the (x, y) components of the multi-point tangential vector, and the numerical value of the direction change are integrated to obtain the tangent direction vector set.

[0078] S103: Based on the tangent direction vector set, solve the principal curvature and secondary curvature values ​​of multiple pixels point by point, and then combine the numerical results according to the feature dimension to establish a multi-machine target feature vector library.

[0079] Based on the tangent direction vector set and combined with image data acquired by multiple UAVs during synchronous acquisition, the principal and secondary curvature values ​​of multiple pixels on the crack target are solved point by point. Since a single image can only provide two-dimensional information, the solution of principal and secondary curvature requires the three-dimensional spatial coordinates of the target. The three-dimensional coordinates are reconstructed by fusing images taken by UAVs from at least two different locations. Specifically, the crack edge points identified in the image of UAV 1, such as P... 10 (150, 320), the corresponding point is found in the synchronized image of UAV 2 using a feature matching algorithm. Assume the coordinates of the corresponding point are P'. 10 (210, 325), given the relative position and attitude parameters of the two drones at the moment of shooting (obtained via RTK-GPS and IMU), the three-dimensional coordinates (X, Y) of the pixel in the real-world coordinate system can be calculated using the principle of triangulation. 10 Y 10 Z 10 For example, the calculated result is (10.52, 25.11, -5.03), in meters. This 3D reconstruction process is repeated for the key pixels on the crack outline to obtain a point cloud sequence in 3D space, as shown in Table 1. This table shows some pixels matched from the images of the two UAVs and their reconstructed 3D space coordinates.

[0080] Table 1. Coordinates of 3D Reconstructed Edge Points:

[0081]

[0082] As shown in Table 1, after obtaining the three-dimensional point set of the crack edge, for each three-dimensional point, for example, the point with index 10... Take its nearest neighbor in the three-dimensional sequence (e.g. (etc.), fitting a local quadratic surface to the nearest points. This fitting process does not directly call the model, but is achieved by solving an overdetermined system of linear equations. The system of equations is based on the coordinates (X, Y, Z) of each nearest point and the quadratic surface equation Z=aX 2 +bY 2 Minimizing the residual of +cXY+dX+eY+f yields the coefficients a, b, c, d, e, and f. According to the principles of differential geometry, the principal curvature k1 and secondary curvature k2 at this point can be calculated using the eigenvalues ​​of the coefficient matrix. For example, after calculation, the point... The principal curvature k1 at point P is 0.85, representing the degree of curvature in the crack extension direction, and the secondary curvature k2 is 0.05, representing the curvature of the surface perpendicular to the crack extension direction. Finally, the numerical results calculated for each key point are spliced ​​and integrated according to the preset feature dimensions to form a high-dimensional feature vector. This vector can contain the following dimensions: three-dimensional spatial coordinates X, Y, Z, principal curvature k1, secondary curvature k2, calculated two-dimensional curvature K, and calculated cumulative angle change A. 10 Its feature vector is [10.52, 25.11, -5.03, 0.85, 0.05, 0.122, 28.77]. By combining the feature vectors of the processed crack key points, a structured database describing the crack target is constructed, namely the multi-machine target feature vector library.

[0083] Please see Figure 3 The specific steps of S2 are as follows:

[0084] S201: Based on the feature vector sequence recorded in the multi-machine target feature vector library, the curvature values ​​under cross-machine and cross-scale conditions are extracted point by point, the rate of change is compared with the preset curvature threshold, and when the rate of change exceeds the threshold, it is marked to generate a curvature anomaly marker set.

[0085] Based on a multi-drone target feature vector library, which stores feature vector sequences in the form [10.52, 25.11, -5.03, 0.85, 0.05, 0.122, 28.77], this library contains crack features obtained and calculated from the perspectives of different drones. The curvature values ​​obtained from different drone platforms and shooting scales are extracted and compared point-by-point. Specifically, the two-dimensional curvature K of a specific point is first extracted from the feature vector, along with the drone number and relative scale parameters at the time of shooting. For example, assuming drone 1 shoots at a standard height of 10 meters, its scale parameters... The scale parameter is 1.0. Due to gusts of wind, the shooting height of drone 2 is reduced to 11 meters, with a scale parameter of 0.9. The shooting height of drone 3 is 10.5 meters, with a scale parameter of 1.1. For the crack point with the same name at index 10 in the feature vector library, the two-dimensional curvature is extracted from the data of drone 1 as K1=0.122, from the data of drone 2 as K2=0.135, and from the data of drone 3 as K3=0.110. Then, the rate of change of curvature values ​​at different scales is calculated. Taking the data at scale 1.0 as the benchmark, the rate of change R of the drone 2 data relative to the benchmark is calculated. 12 =|K2-K1| / K1=|0.135-0.122| / 0.122≈0.1066, calculate the rate of change R of UAV 3 data relative to the baseline. 13=|K3-K1| / K1=|0.110-0.122| / 0.122≈0.0984, and compare the calculated rate of change with a preset curvature threshold T. K For comparison, this threshold is set based on statistical analysis of curvature changes in a large number of normal, continuous crack segments photographed at different scales. The 95th percentile of the rate of change distribution is taken as the threshold. If the statistics show that the calculated curvature change rate of a normal crack does not exceed 0.10 within a height variation range of ±10%, then the curvature threshold T is set. K Set it to 0.10, and set the rate of change R 12 =0.1066 and threshold T K =0.10. Since 0.1066 > 0.10, it is determined that there is a curvature anomaly between the data of UAV 1 and 2 at this point. The index 10 of this point and the UAV number (1, 2) involved are marked, and the rate of change R 13 =0.0984 and threshold T K =0.10 comparison. Since 0.0984≤0.10, no marking is performed. The same rate of change calculation and threshold comparison process is performed on the curvature of the same points in the feature vector library across different UAV data. The index of the points judged as abnormal and the corresponding UAV number are recorded to generate a curvature anomaly marker set.

[0086] S202: Call the curvature anomaly marker set, retrieve the scale parameters of the marker positions one by one, calculate the difference between the multi-scale parameters and the unified reference scale, and then compare it with the correction threshold. The indexes that exceed the threshold are recorded and merged to obtain the scale correction index set.

[0087] The curvature anomaly marker set is invoked, which records entries such as "index 10, UAV pair (1, 2)". The scale parameters of each marked location are retrieved item by item. For the marker at index 10, the scale parameter of UAV 1 is retrieved from the original data record as S1=1.0, and the scale parameter of UAV 2 is S2=0.9. Next, the retrieved multi-scale parameters are compared with a unified reference scale S. base For the difference calculation, the reference scale here is selected as the scale at the design flight altitude, i.e., S. base =1.0, calculate the difference between the scale parameter and the reference scale, and obtain ΔS1=|1.0-1.0|=0 and ΔS2=|0.9-1.0|=0.1. Then, add the difference results to a correction threshold T. S The correction threshold is set based on the positioning accuracy and attitude stability technical indicators of the UAV flight control system. For example, if the UAV's altitude holding accuracy is ±0.2 meters, for a reference flight altitude of 5 meters, the allowable range of scale variation is 0.2 / 5 = 0.04. Therefore, the correction threshold T can be set accordingly. SSet to 0.04, compare the difference ΔS1=0. 0≤0.04, this scale parameter does not need correction. Compare the difference ΔS2=0.1. 0.1>0.04, indicating that the shooting scale of UAV 2 exceeds the normal error range and needs correction. Therefore, record the index 10 of this point. Assume there is another entry in the curvature anomaly marker set, "Index 45, UAV pair (1,3)", whose corresponding scale parameters are S1=1.0 and S3=1.1. The difference calculation results are ΔS1=0 and ΔS3=|1.1-1.0|=0.1. Since ΔS3=0.1>T S =0.04, so index 45 is also recorded. After traversing all entries in the curvature anomaly marker set, the recorded indices, such as 10 and 45, are merged and deduplicated to obtain a final index list, which is the scale correction index set.

[0088] S203: Based on the scale-corrected index set, the probability density distribution of the coordinate vectors of the multi-machine edge pixels corresponding to the index is estimated. A Gaussian mixture model is used to fit the density and perform collaborative weighted localization. The matrix is ​​aggregated to relocate the coordinate vectors and obtain the standardized target dataset.

[0089] Based on the scale-corrected index set, such as a list containing indices [10, 45, ...], the probability density distribution of the coordinate vectors of the multi-drone edge pixels corresponding to each index in the list is estimated. Taking index 10 as an example, the pixel coordinates corresponding to this point are extracted from the original image data of the three drones, namely P1(150, 320) for drone 1, P2(210, 325) for drone 2, and P3(182, 315) for drone 3. These three coordinate points are regarded as an observation sampling with error of the true position of the target. The position is estimated by fitting the density distribution of the sampling points. Specifically, the geometric center (mean) of this set of coordinate points is first calculated as the initial estimated position:

[0090] ;

[0091] Subsequently, based on the calibration parameters of the UAV cameras and real-time attitude data, a cooperative positioning weight is assigned to each observed coordinate. This weight is set according to the camera resolution, the accuracy of the attitude sensor, and the shooting distance. For example, UAV 1, with its high-precision main camera, has a weight W1 = 0.5; UAV 2, with its lower flight altitude and higher resolution, has a weight W2 = 0.4; and UAV 3, with its slightly weaker camera performance and attitude jitter, has a weight W3 = 0.1. The sum of the weights, W1 + W2 + W3 = 1.0. Then, the initial position is weighted and corrected using these weights, and the weighted average coordinate P is calculated. re-located=(X', Y'), where X'=W1·X1+W2·X2+W3·X3=0.5×150+0.4×210+0.1×182=75+84+18.2=177.2, and Y'=W1·Y1+W2·Y2+W3·Y3=0.5×320+0.4×325+0.1×315=160+130+31.5=321.5, resulting in the repositioned coordinate vector (177.2, 321.5). This process essentially uses the idea of ​​Gaussian mixture to fit multiple observation points into a Gaussian distribution centered on the weighted mean, and takes its peak point as the optimal estimate. The same coordinate vector extraction, weighting, and repositioning calculation are performed on each index in the scale correction index set, as shown in Table 2, which summarizes the multi-machine observation coordinates and the final repositioned coordinates of some index points.

[0092] Table 2. Multi-machine observation point coordinate relocation table:

[0093]

[0094] As shown in Table 2, the relocation coordinate vectors (X', Y') obtained from the index points are aggregated into a matrix, that is, the two-dimensional coordinate vectors are organized into an N×2 matrix according to the index order, where N is the number of indexes in the scale-corrected index set. This matrix is ​​the standardized target dataset obtained after multi-machine data fusion and scale correction.

[0095] Please see Figure 4 The specific steps of S3 are as follows:

[0096] S301: Construct a multi-view target region based on a standardized target dataset, perform geometric subdivision operations on the target region and aggregate the boundary point coordinates to form a spatial range, perform averaging operations on the coordinate point set within the range, and generate a sub-block centroid coordinate set.

[0097] Based on a standardized target dataset, the corrected two-dimensional pixel coordinates are first mapped using the calculated three-dimensional spatial information of the points to generate a multi-view target region point cloud describing the spatial morphology of the target crack, composed of higher-precision three-dimensional coordinate points. For example, mapping the corrected two-dimensional coordinates (177.2, 321.5) at index 10 back to three-dimensional space yields updated three-dimensional coordinates (10.53, 25.12, -5.03), in meters. Next, a geometric partitioning operation is performed on the target region point cloud. Specifically, following the index order of the crack points, 20 consecutive three-dimensional coordinate points are divided into an independent sub-block. If the point cloud contains 40 points, it is divided into sub-block 1 (indexes 1-20) and sub-block 2 (indexes 21-40). Then, the coordinates of the boundary points within each sub-block are aggregated to... To determine the spatial extent, for sub-block 1, iterate through its 20 contained 3D coordinate points and find the minimum and maximum values ​​in the X, Y, and Z dimensions. For example, after iteration and calculation, the X coordinate range of sub-block 1 is [10.50, 10.85], the Y coordinate range is [25.05, 25.45], and the Z coordinate range is [-5.10, -5.02]. This cuboid region is the spatial extent of the sub-block. Finally, within the determined spatial extent, average the set of 20 coordinate points within the sub-block, that is, add the X coordinate values ​​of the points and divide by 20. Perform the same operation on the Y and Z coordinates to generate the centroid coordinates of the sub-block. Assuming the sum of the X coordinates of the 20 points in sub-block 1 is 214.60, the sum of the Y coordinates is 506.00, and the sum of the Z coordinates is -101.20, then the centroid coordinates C1 of sub-block 1 are calculated as follows:

[0098] (214.60 / 20, 506.00 / 20, -101.20 / 20) = (10.730, 25.300, -5.060);

[0099] Perform the same subdivision, range determination, and averaging process on sub-block 2. Assume that the calculated centroid coordinates C2 are (11.050, 25.720, -5.030). Arrange the calculated centroid coordinates in the order of sub-blocks to form the sub-block centroid coordinate set.

[0100] S302: Call the sub-block centroid coordinate set, perform difference operation on the same sub-block centroid coordinates in adjacent observation times, and reassemble the difference results into an ordered vector sequence after decomposing the three-dimensional axial components to obtain the centroid coordinate displacement vector set.

[0101] The process calls up the centroid coordinate set of the sub-blocks, which contains coordinates such as C1 (10.730, 25.300, -5.060) and C2 (11.050, 25.720, -5.030). At an adjacent observation time T2 (e.g., T2 = T1 + 0.1 seconds), the entire process is repeated for the same target area to generate a new centroid coordinate set for the sub-blocks. Since the bridge experiences minor vibrations under wind or traffic loads, the centroid coordinates at time T2 will change. Assuming the new centroid coordinates of sub-block 1 and sub-block 2 calculated at time T2 are C'1 (10.732, 25.299, -5.063) and C'2 (11.051, 25.722, -5.031), respectively, a difference operation is performed on the centroid coordinates of the same sub-blocks within two adjacent observation times to obtain the displacement of the centroid of each sub-block. For sub-block 1, its displacement difference vector... Calculated using C'1-C1, i.e.:

[0102] (10.732-10.730, 25.299-25.300, -5.063-(-5.060))=(0.002, -0.001, -0.003), in meters;

[0103] For sub-block 2, its displacement difference vector Calculated using C'2-C2, i.e.:

[0104] (11.051-11.050, 25.722-25.720, -5.031-(-5.030))=(0.001, 0.002, -0.001), in meters;

[0105] Then, each difference operation result, i.e., the displacement vector, is decomposed into its three-dimensional axial components. It is decomposed into X component ΔX1=0.002, Y component ΔY1=-0.001, Z component ΔZ1=-0.003, and vector. It is decomposed into ΔX2=0.001, ΔY2=0.002, ΔZ2=-0.001. Finally, the components are recombined in the order of the sub-block index and the fixed order of X, Y, Z to form an ordered vector sequence [0.002, -0.001, -0.003, 0.001, 0.002, -0.001, ...]. This complete sequence is the centroid coordinate displacement vector set.

[0106] S303: Based on the centroid coordinate displacement vector set, perform aggregate statistics on the direction angle and amplitude parameters of the vectors, and arrange the statistical results in order into a matrix structure to obtain the spatial positioning data of the UAV cluster.

[0107] Based on the set of centroid coordinate displacement vectors, where , For each vector, aggregate statistics are performed on its direction angle and magnitude parameters. First, the magnitude of each vector is calculated, where the magnitude represents the magnitude of the displacement. Its amplitude:

[0108] rice;

[0109] for Its amplitude:

[0110] rice;

[0111] Next, calculate the direction angle of each vector, that is, the angle between the vector and the positive directions of the X, Y, and Z coordinate axes. Its orientation angles are α1 = arccos(0.002 / 0.00374) ≈ 57.7°, β1 = arccos(-0.001 / 0.00374) ≈ 105.5°, and γ1 = arccos(-0.003 / 0.00374) ≈ 143.3°. Performing the same calculations, we obtain α2=arccos(0.001 / 0.00245)≈65.9°, β2=arccos(0.002 / 0.00245)≈35.3°, and γ2=arccos(-0.001 / 0.00245)≈114.1°. After calculating the magnitude and orientation angle of the displacement vector of the sub-block, the statistical results are arranged in the index order of the sub-blocks and organized into a matrix structure, as shown in Table 3.

[0112] Table 3: Statistics on centroid displacement

[0113]

[0114] As shown in Table 3, the table clearly shows the three-dimensional spatial displacement characteristics of different segments of the crack within a specified time interval. By integrating the data stream of this matrix structure, we can obtain UAV swarm spatial positioning data for characterizing the dynamic response of the target structure.

[0115] Please see Figure 5 The specific steps of S4 are as follows:

[0116] S401: Based on the spatial positioning data of the UAV cluster, the three-dimensional parameters of the coordinate points in the data are numerically weighted according to the weight coefficients. The weighted coordinate set is compared with the sum parameter matrix and then a normalization operation is performed to generate the correction vector value.

[0117] Based on UAV swarm spatial positioning data, the three-dimensional coordinates of the centroids of the corresponding sub-blocks, specifically C1 (10.730, 25.300, -5.060) and C2 (11.050, 25.720, -5.030) at time T1, are numerically weighted according to a set of weighting coefficients. These weighting coefficients are set with reference to the stability of the sub-block displacement; that is, sub-blocks with lower displacement amplitudes are considered to have more stable observation data and are therefore assigned higher weights. The weight W... i The specific calculation process is as follows: take the displacement amplitude M of the sub-block. i The reciprocal of the first value is divided by the sum of the reciprocals of the sub-block displacement amplitudes. For sub-blocks 1 and 2, their displacement amplitudes are M1 = 0.00374 meters and M2 = 0.00245 meters, respectively. Therefore, the weights are:

[0118] W1 = (1 / 0.00374) / ((1 / 0.00374)+(1 / 0.00245))≈267.38 / (267.38+408.16)≈0.396, and similarly W2≈0.604;

[0119] Applying the weights to the centroid coordinates at time T1 yields a weighted set of coordinates, the weighted average of which is:

[0120] ;

[0121] Performing the same weighted operation on C'1 (10.732, 25.299, -5.063) and C'2 (11.051, 25.722, -5.031) at time T2 yields:

[0122] ;

[0123] Next, we define a summation parameter matrix, which here refers to the global displacement vector defined by the weighted average displacement of all sub-blocks. The actual displacement vector of the sub-block, such as This is compared with the global displacement vector, specifically by performing vector subtraction to obtain the relative displacement vector. Finally, a normalization operation is performed on this relative displacement vector, which involves dividing its components by the magnitude of the vector. This yields the unit vector (0, -0.894, -0.447), which is related to the relative displacement vector. Together, they constitute the correction vector value of this sub-block.

[0124] S402: Call the correction vector value, perform translation compensation calculations on the horizontal and vertical parameters in the centroid coordinate matrix of the sub-block respectively, update the calculated coordinate points in parallel, and obtain the translation compensation coordinate set;

[0125] Call the correction vector value, specifically the relative displacement vector of the sub-block, for example, the value of sub-block 1. and sub-block 2 For the horizontal and vertical parameters in the centroid coordinate matrix of the sub-block, namely the three-dimensional coordinates of the centroid C'1 (10.732, 25.299, -5.063) and C'2 (11.051, 25.722, -5.031) observed at time T2, translation compensation calculations are performed. The calculation is performed by subtracting the corresponding relative displacement vector from the observed coordinates at time T2. This operation filters out the local relative motion components of each sub-block relative to the overall structure. For sub-block 1, its compensated coordinate C”1 is calculated as (10.732, 25.299, -5.063) - (0, -0.002, -0. 001)=(10.732, 25.301, -5.062). For sub-block 2, its compensated coordinate C”2 is calculated as (11.051, 25.722, -5.031)-(-0.001, 0.001, 0.001)=(11.052, 25.721, -5.032). The calculated coordinate points (10.732, 25.301, -5.062) and (11.052, 25.721, -5.032) are updated in parallel, that is, the compensation calculation is completed independently for each sub-block, and the compensated coordinate results are collected to obtain the translation compensation coordinate set.

[0126] S403: Based on the translation compensation coordinate set, the predicted and observed values ​​of time series coordinate points are differentially analyzed. The Kalman filter algorithm is used to input the differential result into the state transition matrix and is combined with the observation residual vector for correction to obtain low-altitude target tracking trajectory data.

[0127] Based on the translation-compensated coordinate set, such as the coordinates C”1 (10.732, 25.301, -5.062) at time T2, the predicted and observed values ​​of this time series coordinate point are differencing. First, based on the state of the previous time T1 (position C1 (10.730, 25.300, -5.060) and the initial velocity estimated from the displacement, To generate the predicted value at time T2, the predicted location is:

[0128] ;

[0129] Then, the predicted value is compared with the observed value at time T2, i.e., the coordinates C”1 after translation compensation, to obtain the observation residual vector:

[0130] ;

[0131] Next, the difference result, i.e., the observation residual, is used to correct the predicted state. Specifically, a state transition process is first set up, assuming the velocity remains constant over a short period. Simultaneously, an observation process is set up, where the position can be directly measured. Based on the ratio between the uncertainty of the predicted state and the uncertainty of the observation noise, a gain coefficient is calculated. Here, the gain coefficient for position correction is set as K. p =0.8, the gain coefficient for speed correction is K v =0.5, combining the observed residual vector with the gain coefficient, the predicted state is corrected, and the corrected position is:

[0132] ;

[0133] Corrected speed:

[0134] ;

[0135] The final position coordinates calculated at each moment are concatenated, as shown in Table 4.

[0136] Table 4 Target Tracking Trajectory Data:

[0137]

[0138] As shown in Table 4, the table lists the final three-dimensional coordinates of sub-block 1 at three consecutive timestamps. This time series table, consisting of a series of corrected coordinate points, is the final low-altitude target tracking trajectory data.

[0139] Please see Figure 6 The specific steps of S5 are as follows:

[0140] S501: Based on low-altitude target tracking trajectory data and preset flight mission parameters, calculate the Euclidean distance between the trajectory data position coordinates and mission track points one by one, compare the difference with the matching degree threshold, record the coordinate index below the threshold and perform time serialization encoding to generate a trajectory matching index sequence.

[0141] Based on low-altitude target tracking trajectory data, and combined with a pre-set flight mission parameter file containing the three-dimensional coordinates of planned waypoints (where WP1=(10.731, 25.300, -5.061) and WP2=(10.733, 25.300, -5.063), the Euclidean distance between each position coordinate in the trajectory data and the mission waypoints is calculated one by one. Taking the trajectory point P at time T2 as an example... T2 Taking (10.7320, 25.3006, -5.0622) as an example, the distance d1 between it and the waypoint WP1 is calculated as follows:

[0142] rice;

[0143] The distance d2 between it and the waypoint WP2 is calculated as follows:

[0144] rice;

[0145] The calculated distance difference is compared with a preset matching threshold T. d For comparison, this threshold is set based on the positioning accuracy indicators of the UAV navigation system. If the RTK positioning module used by the UAV has a nominal horizontal accuracy of 1 cm and a vertical accuracy of 2 cm, then the overall positioning error is approximately... To account for system latency and environmental interference, the matching degree threshold T can be set to [value]. d Set to 0.03 meters, and compare the distance d1 = 0.00167 meters with T. d Comparing, 0.00167 < 0.03, we compare the distance d2 = 0.00141 meters with T. d Comparing the two distances, 0.00141 < 0.03, both distances are less than the threshold. Therefore, the track point with the smaller distance is selected as the matching point, and thus P... T2 Matching with WP2, record the index of the trajectory point, i.e. its timestamp T2, and the index of the matching track point 2. Perform the same distance calculation and threshold comparison process on the trajectory points. Arrange the coordinate point indices below the threshold according to their own time order and pair them with the matching track point indices to form an encoded sequence, such as [(T1, 1), (T2, 2), (T3, 2), ...]. This sequence is the trajectory matching index sequence.

[0146] S502: Call the trajectory matching index sequence, extract the corresponding velocity vector from the multi-machine trajectory dataset, split the velocity vectors of the same time slice into direction angle and magnitude and calculate the difference, compare the difference with the anomaly detection threshold, mark the time slice index that exceeds the threshold, and obtain the abnormal time slice set.

[0147] The trajectory matching index sequence is invoked, for example, [(T1, 1), (T2, 2), (T3, 2), ...]. The actual velocity vector corresponding to each timestamp index is extracted from the multi-aircraft trajectory dataset, and the planned velocity vector corresponding to the matching track point index is extracted from the mission parameter file. Taking time T2 as an example, its corresponding actual velocity vector is calculated. m / s, and the planned velocity vector of its matched waypoint WP2 is meters per second, decompose the two velocity vectors in the same time slice into direction angle and magnitude, the magnitude of the actual velocity. meters per second, the modulus of the planned speed The speed is m / s. Next, the difference between the two is calculated. The difference in magnitude is |0.0320 - 0.0224| = 0.0096 m / s. The difference in direction angle is obtained by calculating the angle θ between the two vectors.

[0148] ;

[0149] Then θ≈arccos(0.9765)≈12.4°. The difference is compared with an anomaly detection threshold. This threshold is set with reference to the response performance of the UAV's power system and the tracking error tolerance of the flight control law. Velocity modulus errors within 15% and direction errors within 15% are considered normal. Therefore, the modulus threshold is set accordingly. m / s, direction angle threshold T a =15°, the calculated modulus difference of 0.0096 is compared with T m Comparing, 0.0096 > 0.0034, the difference in direction angle of 12.4° is compared with T. a For comparison, 12.4°≤15°. Since the difference in modulus exceeds the threshold, the time slice index T2 is marked. After traversing the trajectory matching index, the marked time slice indices are collected together to obtain the abnormal time slice set.

[0150] S503: Based on the abnormal time slice set, call the task parameter waypoints, reassign the velocity vector and position coordinates to form a new waypoint sequence, and synchronize it with the original flight state parameters in the time dimension to obtain the low-altitude economic UAV operation status result;

[0151] Based on the set of abnormal time slices, such as a list containing timestamps T2 and T5, the waypoint information in the preset flight mission parameters is retrieved, and the state is corrected for each abnormal time point. Taking time T2 as an example, the actual position of the UAV at this time is P. T2 (10.7320, 25.3006, -5.0622), but its speed is judged to be abnormal. At this time, a reallocated velocity vector is concatenated with the position coordinates at this moment to form a new temporary trackpoint. The logic for generating this reallocated velocity vector is to adopt the planned velocity of the next planned trackpoint WP3. However, its modulus length is reduced by 30% as a safety buffer. m / s, then the new velocity vector m / s, therefore, the new waypoint generated for time T2 is WP' T2Its position coordinates are (10.7320, 25.3006, -5.0622), and its velocity vector is (0.007, 0, -0.014). This new waypoint sequence will be used to replace the original waypoint starting from time T2. Next, this new waypoint information will be synchronously encoded with the original flight state parameters at that time, such as the UAV's attitude angles (roll 0.1°, pitch -0.2°) and battery voltage (14.8 volts), in the time dimension, that is, a data structure containing key state information will be created, as shown in Table 5.

[0152] Table 5. UAV Operation Status Table:

[0153]

[0154] As shown in Table 5, the table records the complete operating status of the UAV after correction in the abnormal time slice T2, as well as the planned status in the next normal time slice T3. After integrating the status data at each time, the operating status result of the low-altitude economic UAV is obtained.

[0155] Please see Figure 7 A low-altitude economic unmanned aerial vehicle (UAV) data processing system, comprising:

[0156] The image curvature extraction module is configured to simultaneously acquire a distributed target image sequence through multiple airborne cameras, perform parallel curvature extraction processing on the target edge pixels in the distributed target image sequence, construct a local tangent direction vector, calculate the principal curvature value and the secondary curvature value, and output a multi-camera target feature vector library.

[0157] The multi-machine feature modeling module is configured to perform threshold judgment on the rate of change of curvature values ​​under cross-machine and cross-scale conditions based on the multi-machine target feature vector library. When the rate of change exceeds the preset threshold range, a distributed scale correction mechanism is triggered, and a Gaussian mixture model is used to perform multi-machine edge pixel collaborative relocation and output a standardized target dataset.

[0158] The target region segmentation module is configured to construct a multi-view target region based on the standardized target dataset, divide the multi-view target region into multiple geometric sub-blocks and extract the centroid coordinates of the sub-blocks, calculate the centroid coordinate displacement vector, perform direction and amplitude statistical analysis, and output UAV cluster spatial positioning data.

[0159] The spatial positioning compensation module is configured to call the spatial positioning data of the UAV cluster to perform weighted mean normalization operation, extract the correction vector value and perform parallel translation compensation on the centroid coordinates of the sub-block, use the Kalman filter algorithm for filtering, and output low-altitude target tracking trajectory data.

[0160] The trajectory fusion determination module is configured to determine the matching degree threshold based on the low-altitude target tracking trajectory data and preset flight mission parameters, perform fusion analysis and anomaly detection on multi-aircraft trajectory data, trigger an alarm mechanism and replan the flight path when the preset threshold is exceeded, and output the low-altitude economic UAV operation status result.

[0161] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A data processing method for low-altitude economic unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: S1: Simultaneously acquire distributed target image sequences through multiple airborne cameras, perform parallel curvature extraction processing on target edge pixels in the distributed target image sequences, construct local tangent direction vectors, calculate principal curvature and secondary curvature values, and output a multi-camera target feature vector library; S2: Based on the multi-machine target feature vector library, the rate of change of curvature values ​​under cross-machine and cross-scale conditions is judged by a threshold. When it exceeds the preset threshold range, a distributed scale correction mechanism is triggered. A Gaussian mixture model is used to perform multi-machine edge pixel collaborative relocation and output a standardized target dataset. S3: Based on the standardized target dataset, construct a multi-view target region, divide the multi-view target region into multiple geometric sub-blocks and extract the centroid coordinates of the sub-blocks, calculate the centroid coordinate displacement vector, perform direction and amplitude statistical analysis, and output UAV cluster spatial positioning data. S4: Call the spatial positioning data of the UAV cluster to perform weighted mean normalization operation, extract the correction vector value and perform parallel translation compensation on the centroid coordinates of the sub-block, use the Kalman filter algorithm for filtering processing, and output the low-altitude target tracking trajectory data. The multi-aircraft target feature vector library includes curvature distribution features, boundary direction features, and shape contour features. The standardized target dataset includes scale-uniform data, coordinate-normalized data, and pixel-consistent data. The UAV swarm spatial positioning data includes regional geometric features, centroid distribution features, and displacement vector features. The low-altitude target tracking trajectory data includes direction sequences, amplitude sequences, and time sequences.

2. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Simultaneously acquire distributed target image sequences through multiple airborne cameras, perform point-by-point geometric curvature extraction on the edge pixels of the target region in the distributed target image sequence, and arrange the curvature values ​​in index order to generate a curvature sequence set; S102: Call the curvature sequence set, vectorize the tangential direction of adjacent pixels, numerically calculate the angle difference and direction change between the tangential vectors, and encode the difference results by index to obtain the tangential direction vector set; S103: Based on the tangent direction vector set, solve the principal curvature and secondary curvature values ​​of multiple pixels point by point, and then combine the numerical results according to the feature dimension to establish a multi-machine target feature vector library.

3. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 2, characterized in that, The specific steps of S2 are as follows: S201: Based on the feature vector sequence recorded in the multi-machine target feature vector library, extract the curvature values ​​point by point under cross-machine and cross-scale conditions, compare the rate of change with a preset curvature threshold, and mark when the rate of change exceeds the threshold to generate a curvature anomaly marker set. S202: Call the curvature anomaly marker set, retrieve the scale parameters of the marker positions one by one, calculate the difference between the multi-scale parameters and the unified reference scale, and then compare it with the correction threshold. Record and merge the indexes that exceed the threshold to obtain the scale correction index set. S203: Based on the scale-corrected index set, perform probability density distribution estimation on the coordinate vectors of the multi-machine edge pixels corresponding to the index, use a Gaussian mixture model to fit the density and perform collaborative weighted localization, and aggregate the coordinate vectors to obtain a standardized target dataset.

4. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 3, characterized in that, The specific steps for S3 are as follows: S301: Based on the standardized target dataset, construct a multi-view target region, perform geometric subdivision operation on the target region and aggregate the boundary point coordinates to form a spatial range, perform average operation on the coordinate point set within the range, and generate a sub-block centroid coordinate set. S302: Call the sub-block centroid coordinate set, perform difference operation on the same sub-block centroid coordinates in adjacent observation times, and reassemble the difference results into an ordered vector sequence after decomposing the three-dimensional axial components to obtain the centroid coordinate displacement vector set; S303: Based on the centroid coordinate displacement vector set, perform aggregate statistics on the direction angle and amplitude parameters of the vectors, and arrange the statistical results in order into a matrix structure to obtain the spatial positioning data of the UAV cluster.

5. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 4, characterized in that, The specific steps of S4 are as follows: S401: Based on the spatial positioning data of the UAV cluster, the three-dimensional parameters of the coordinate points in the data are numerically weighted according to the weight coefficients. The weighted coordinate set is compared with the sum parameter matrix and then a normalization operation is performed to generate a correction vector value. The sum parameter matrix is ​​a global displacement vector defined by the weighted average displacement of all sub-blocks; S402: Call the corrected vector value, perform translation compensation calculations on the horizontal and vertical parameters in the centroid coordinate matrix of the sub-block respectively, update the calculated coordinate points in parallel, and obtain the translation compensation coordinate set; S403: Based on the translation compensation coordinate set, the predicted and observed values ​​of the time series coordinate points are differentially analyzed. The Kalman filter algorithm is used to input the differential result into the state transition matrix and is combined with the observation residual vector for correction to obtain low-altitude target tracking trajectory data.

6. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 1, characterized in that, It also includes the following steps: S5: Based on the low-altitude target tracking trajectory data and preset flight mission parameters, a matching degree threshold is determined, multi-aircraft trajectory data is fused and analyzed and anomalies are detected, and when the preset threshold is exceeded, an alarm mechanism is triggered and the flight path is replanned, and the low-altitude economic UAV operation status result is output. The operational status results of the low-altitude economic UAV include mission matching degree, abnormal alarm information, and flight path planning scheme.

7. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 6, characterized in that, The specific steps of S5 are as follows: S501: Based on the low-altitude target tracking trajectory data and preset flight mission parameters, calculate the Euclidean distance between the trajectory data position coordinates and mission track points one by one, compare the difference with the matching degree threshold, record the coordinate index below the threshold and perform time serialization encoding to generate a trajectory matching index sequence. S502: Call the trajectory matching index sequence, extract the corresponding velocity vector from the multi-machine trajectory dataset, split the velocity vectors of the same time slice into direction angle and magnitude and calculate the difference, compare the difference with the anomaly detection threshold, mark the time slice index that exceeds the threshold, and obtain the abnormal time slice set. S503: Based on the set of abnormal time slices, call the task parameter waypoints, reassemble the reallocated velocity vectors and position coordinates to form a new waypoint sequence, and synchronize it with the original flight state parameters in the time dimension to obtain the low-altitude economic UAV operation status result.

8. The low-altitude economic unmanned aerial vehicle (UAV) data processing method according to claim 7, characterized in that, The matching degree threshold is determined based on the UAV navigation and positioning accuracy and the allowable deviation of track tracking. The anomaly detection threshold is determined based on the fluctuation range of the UAV's speed modulus and the deviation range of its azimuth angle.

9. A low-altitude economic unmanned aerial vehicle (UAV) data processing system, used to implement the low-altitude economic UAV data processing method according to any one of claims 1-8, characterized in that, include: The image curvature extraction module is configured to simultaneously acquire a distributed target image sequence through multiple airborne cameras, perform parallel curvature extraction processing on the target edge pixels in the distributed target image sequence, construct a local tangent direction vector, calculate the principal curvature value and the secondary curvature value, and output a multi-camera target feature vector library. The multi-machine feature modeling module is configured to perform threshold judgment on the rate of change of curvature values ​​under cross-machine and cross-scale conditions based on the multi-machine target feature vector library. When the rate of change exceeds the preset threshold range, a distributed scale correction mechanism is triggered, and a Gaussian mixture model is used to perform multi-machine edge pixel collaborative relocation and output a standardized target dataset. The target region segmentation module is configured to construct a multi-view target region based on the standardized target dataset, divide the multi-view target region into multiple geometric sub-blocks and extract the centroid coordinates of the sub-blocks, calculate the centroid coordinate displacement vector, perform direction and amplitude statistical analysis, and output UAV cluster spatial positioning data. The spatial positioning compensation module is configured to call the spatial positioning data of the UAV cluster to perform weighted mean normalization operation, extract the correction vector value and perform parallel translation compensation on the centroid coordinates of the sub-block, use the Kalman filter algorithm for filtering, and output low-altitude target tracking trajectory data. The trajectory fusion determination module is configured to determine the matching degree threshold based on the low-altitude target tracking trajectory data and preset flight mission parameters, perform fusion analysis and anomaly detection on multi-aircraft trajectory data, trigger an alarm mechanism and replan the flight path when the preset threshold is exceeded, and output the low-altitude economic UAV operation status result.

Images